A Two-Channel Short Text Intent Recognition Model Combining CNN and BiGRU

In the field of short text intention recognition, convolutional neural networks (CNNS) have attracted much attention because of their excellent performance in local information extraction. However, because it is difficult to capture the global features of short text corpus, it has some limitations. To solve this problem, this paper proposes a dual-channel short text intent recognition model combining the advantages of TextCNN and BiGRU-att, which uses local features and global features to better identify short text intent and make up for the deficiency of the model for the overall text features. The AB-CNN-BGRU-att model first uses the ALBERT multilayer bidirectional Transformer structure to vectorize the input text, and then feeds the vectors into the TextCNN and BiGRU network models respectively to obtain local and global features. The two features are fused, and the intent label is obtained through the full connection layer and input Softmax function. The experimental results show the feasibility of the model.